faster-whisper
A fast reimplementation of Whisper on CTranslate2, with much lower latency and memory use.
faster-whisper reimplements Whisper inference on the CTranslate2 engine. The result is the same model with several times the speed and a smaller memory footprint: the same transcript, far cheaper to produce.
It is the usual production answer to "Whisper is too slow": a drop-in upgrade that keeps accuracy while making self-hosted transcription practical at scale.
Where it fits
The default when you want Whisper-quality transcription self-hosted, but the reference implementation is too slow or memory-hungry.
Where it does not fit
A faster runtime, not new capability: it does not add streaming or diarisation on its own.
Alternatives to faster-whisper
The closest options on the map, and where each fits instead.
Pairs with
Commonly used alongside it.